fidoptPaired head CT and MR, MR to CT
Contrast-enhanced T1-weighted MR of the head of a patient before neurosurgery and the thin-slice CT to synthesise from it, both put on one 0.5 mm grid by the provider (A Paired Head CT-MRI Dataset for Cross-Modality Image Synthesis, The First Affiliated Hospital, Sun Yat-sen University).
Read before comparing (8)
- Source. The contrast-enhanced T1-weighted 3D MPRAGE (CE-T1W.nii.gz) on Siemens Verio, Prisma, Vida or Spectra scanners, as delivered: resampled by the provider from 0.38 to 1 mm pixels onto the 0.5 mm grid it shares with the CT, floats without unit, 0 where the MR did not reach. The pituitary tumours also have a T1 before contrast, not used, so every case has the same input.
- Target. CT-BRAIN.nii.gz, the soft-kernel reconstruction (FC26, Hr40, FC30, UB and others) of a 0.5 or 0.625 mm head CT on Toshiba, Siemens, Philips or GE scanners, as delivered: in Hounsfield units, -1024 outside the scan, resampled by the provider from 0.36 to 0.79 mm pixels to 0.5 mm, so its values are interpolated. The sharp-kernel reconstruction of the same scan (CT-BONE) is not used.
- Alignment. The provider gives the two images on one grid and does not say how it registered them. On the 12 test cases a rigid registration on mutual information from identity (it brings a known 3 mm, 3 degree move back to within 0.8 mm) still moves the MR by 0.2 to 2.0 mm (median 0.6 mm), almost all along the head-foot axis, and rotates it 0.6 degrees at most; pituitary-tumor-sub-39 is the worst. The CT's bone lies on the MR's skull in the overlays. The provider's alignment is kept: what it leaves misaligned counts as error for every method.
- Face. Both images are defaced, the face cut away in the same place.
- Mask. The head: the CT above -500 HU, its largest connected part, the holes of each axial slice filled (sinuses, airways), inside the MR's reach. It holds 2.5 to 3.5 litres on the test cases.
- Baseline. Inside the mask on the 12 test cases, with KonfAI's criteria at CT's range of 4095, a constant at the target's median there (28 to 39 HU) scores an MAE of 214 HU (176 to 250), a PSNR of 19.2 dB (17.9 to 21.1) and an SSIM of 0.50 (0.43 to 0.55). The MR is not in Hounsfield units and gives no floor.
- Exposure. The record has been public with its CTs since October 2025: a network trained on it may have seen these patients.
- Overlap. No other dataset of the corpus comes from Sun Yat-sen University or names Zenodo 17486320; SynthRAD 2023 and 2025 are from Dutch and German centres, PituitaryTumourMR from London, the vestibular schwannoma sets from UK and US centres, BraTS_MEN's meningiomas from six US centres. participants.xlsx gives no two rows the same metadata.
No method has a result on every case of this dataset yet.
Under a licence that is not open
Methods whose code or weights are released for non-commercial use, or under share-alike or custom terms: shown for reference, ranked nowhere.
| Accuracy | Cost | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| # | Method | MAE | PSNR dB | SSIM | Anatomy Dice | Anatomy HD95 mm | Time s | RAM GB | GPU GB | Machine | Images |
| KonfAI synthesis app · TotalSynth MR to CT, KonfAI appGPUapptrained on CC-BY-NC-4.0 dataolder run | 123.7 ±10.7 | 24.58 ±0.74 | 0.687 ±0.029 | 0.758 ±0.112 | 5.08 ±1.82 | 22 | 4.2 | 5.0 | Workstation | ||
| nnsyn · MR to CT, head and neckGPUapptrained on CC-BY-NC-4.0 dataolder run | 142.7 ±19.4 | 23.18 ±1.26 | 0.650 ±0.036 | 0.672 ±0.147 | 5.94 ±2.15 | 7.5 | 2.7 | Workstation | |||
MAE: mean absolute error of a synthetic image against its target, inside the body, in the image's units. PSNR: the same by signal-to-noise ratio, higher is better. SSIM: the same by structural similarity, 1 is perfect. Anatomy Dice and HD95: TotalSegmentator run on the synthetic and on the target image, the structures compared; shown, not ranked.
Case by case
| MAE | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 |
|---|
Click a score to look at that result: the target image, and over it the method's synthetic image. Underlined scores have their image; for the others you see the case without the result.
About the data
- Test cases
- 12 test patients drawn with seed 2026 among the record's 120, by population in proportion (pituitary tumour 60, meningioma, vestibular schwannoma and cerebrovascular disease 20 each), then 10 development patients, the last of the same seeded order, each population in proportion. The record gives no split.
- Truth
- target image as released the patient's pre-operative thin-slice CT, soft-kernel reconstruction, in Hounsfield units, on the grid the provider put both images on
- Licence
- CC BY-NC 4.0
10 cases for tuning, never tested
pituitary-tumor-sub-54, pituitary-tumor-sub-46, meningioma-sub-06, vestibular-schwannoma-sub-10, cerebrovascular-disease-sub-05, pituitary-tumor-sub-52, pituitary-tumor-sub-44, pituitary-tumor-sub-14, meningioma-sub-15, vestibular-schwannoma-sub-11